Eka Robotics raises $150M to develop physics-based robot models

Securing capital for dexterity

Eka Robotics recently announced it has raised $150 million in Series A funding to scale its technology and expand customer deployments. The company, which was founded in 2024, is based in Cambridge, Massachusetts. This investment comes from a group of prominent investors including Andreessen Horowitz (a16z) and F-Prime, a fund affiliated with Fidelity Investments.

The funding round also included continued support from seed investors Obvious Ventures and E14 Fund. Other participants in the round included MFV Partners, Bison Ventures, Glasswing Ventures, Samsung Next, and several Fortune 500 strategics. These backers are looking to level up dexterous manipulation, which is the ability of a robot to handle objects with precision and skill.

Investors are betting on Eka's specific approach to solving the hardest problems in robotics. As Raghu Raghuram of a16z noted, the company's model is the first he has seen that truly solves manipulation with the necessary generality and throughput for real-world use. The goal is to push the frontier of robotics research while simultaneously scaling commercial operations.

Moving beyond human mimicry

Eka Robotics believes that for robots to be truly general-purpose, they must master the laws of physics rather than simply mimicking human data. Many current systems rely on watching humans perform a task and trying to copy those movements exactly. Eka argues that this approach limits a robot's ability to handle changing conditions or perform useful work every day in a dynamic environment.

Instead of imitation, the company is building a vision-force-action (VFA) model. This model allows robots to build an intuition for tasks by jointly reasoning about what they see and the forces they feel. By understanding physics, a robot can adapt to new objects or environments that were not part of its original training set. This provides the versatility needed for complicated work in the real world.

The company asserts that this physics-based reasoning is what allows their robots to handle a wide variety of objects outside of their training distribution. For example, robots using this AI can perform roughly 1,500 operations per hour. They are capable of performing delicate tasks such as inserting wires or screwing in a light bulb, which require a nuanced understanding of how objects interact with one another.

Scaling through simulation

To train these complex models, Eka Robotics utilizes simulation learning to overcome the limitations of physical training. In a digital simulation, the company can generate thousands of hours of robot experience in just minutes. This environment allows them to vary objects, physical conditions, and surroundings at a speed that would be impossible to achieve in the physical world.

This method addresses the typical sim-to-real barrier, which is the difficulty of transferring knowledge from a virtual environment to a physical one. Eka claims that their VFA model demonstrates this barrier is falling. By using simulation to provide a breadth of experience, the company can create a more performant and capital-efficient path toward a foundation model for manipulation.

The startup emphasizes that reinforcement learning at scale is the key to buying performance. By exposing the AI to a massive variety of scenarios in simulation, the robot learns to generalize its actions. This allows the robot to remain effective even when faced with unexpected variables, making the transition from a virtual test to a factory floor much more reliable.

Impact on industrial work

The value of a robot on a factory floor is often measured by two specific metrics: its success rate and its speed over thousands of repetitions. Shyam Kamadolli of Fine Structure Ventures noted that current approaches to generalization simply will not reach these goals. He argues that reaching the next tier of capability requires a deep grasp of physical principles and a move away from relying on human demonstrations.

Eka's technology targets the most difficult problems in robotics by focusing on dexterous manipulation. By providing a robot with an intuition for physics, the company aims to make robots more useful in environments where conditions are constantly changing. This means robots can be deployed in more complex roles that require them to interact with a variety of different tools and materials.

As the company scales its customer deployments, the focus remains on delivering useful work every day. The goal is to move away from robots that can only do one specific task perfectly. Instead, Eka wants to create robots that can handle a wide range of tasks with the same level of reliability, making them more viable for diverse industrial applications across different sectors.